Robotic piece-picking: automating the supply chain
October 17, 2018
By Vince Martinelli
October 17, 2018 – Traditional material handling processes focused on moving products by the case or pallet from distribution centres to brick-and-mortar retail stores, where customers would do their shopping. Today, with the convenience of e-commerce, people are shopping over the internet more frequently and buying fewer items at a time. E-commerce has turned the traditional material handling and logistics landscape on its head, compelling us to rethink the overall approach to intralogistics operations within the four walls of the fulfillment centre.
The problem boils down to pieces and people. The “pieces” part of the problem comes from this shift towards shipping individual items, driven by surging online order volume, with double-digit growth spanning millions of SKUs. How can automation cope with this flood of individual items? The “people” part of the problem comes from tight labour markets in many regions, shrinking labour pools in others and a tendency for people to avoid sticking with mundane pick-and-place task-based roles. Even with automated storage and retrieval (ASRS) systems, intelligent conveyors and sorting systems, people play a key role in handling items as they move through the facility. How will retailers grow and sustain outbound volume with the human resource challenges they face?
The solution? Automation of simple, menial warehouse tasks by way of robotic piece-picking, designed for the new realities of e-commerce and for integration with a wide range of warehouse systems.
The adoption of robotic piece-picking for supply chain is still in an early stage, but many retailers and fulfillment centres are already testing and integrating this new technology into their warehouses in order to stay ahead of the competition.
The challenge for someone trying to pick the best option is deciphering the techno-babble and identifying a solution that meets the needs of their business in terms of three key requirements: range, rate and reliability, or simply, the 3Rs.
Range of items
Range of items is a measure of the SKU variety and order diversity in your operation.
SKU count alone is not a perfect measure of variety in regard to size and shape of items. There are many identically packaged tubes of toothpaste, bottles of shampoo and cereal boxes, for example. Even so, it’s likely that the more SKUs your company stocks, the greater the variety of these classes of items and the greater the size options offered. People deal with this variety easily, but traditionally robotics have only been used when the range of items is very low, such as in a manufacturing production line.
But the sheer variety of items is not the only challenge. With demand shifting to direct-to-consumer purchases, your customer order profile may only have between one and two units on average, compared to traditional store-restocking orders that may have 20 or even 100 order lines. Even with some small sets of faster-moving (more popular) products, order diversity is high, especially when trying to process orders quickly to meet same- or next-day deliveries limits batch size. Combining these factors means that an associate doing ASRS tending or sorter induction may not see the same item twice in a row very often through the day.
Does a robotic solution need to support 100 per cent of your products to have a reasonable payback? Probably not. Bicycle tires and toothpaste, for example, do not often flow through the same storage and processing steps in a warehouse. They are unlikely to be packaged together and will normally ship in separate boxes. A toaster oven may ship in its own box and always be processed as a single item. That said, the wider the range of items that your robot workcell can pick, the better. It simplifies material flow, and gives you flexibility to consider using such a system at one or more pain points in the operation.
Rate of picking
Robotic piece-picking has gotten very fast. In a continuous cycle where a robot transfers items one at a time between two adjacent totes, rates of 1,000 units per hour or more can be sustained. Commercial off-the-shelf (COTS) robot arms provide options for rapid picking, and task-specific path-planning algorithms improve efficiency. Intelligent solutions that integrate vision and smart grasping hardware with software intelligence are quick to identify items, select a motion path and execute the pick.
Ask prospective robotic piece-picking partners about benchmark performance measures for items similar to what you intend to run through the workcell. These results should be supported by both lab and field data, accounting for variations in solution design. The supplier should be willing to discuss trade-offs in order to optimize the system for your operation, whether an existing facility or a greenfield design. They may recommend changes to inbound procedures, including how items are presented in totes and how they are packaged, to make it simpler for an automated system to perceive and grasp items. Robotic piece-picking can meet the throughput requirements for many typical DC and FC processes – and it is improving rapidly.
Another consideration regarding rates and throughput is how quickly robotic piece-picking can be deployed and integrated into your systems. The best systems can be deployed in roughly a day by a single technician and feature simple interfaces. As the installed experience base continues to evolve and standardize, these systems will become even more plug-and-play. There is no need to wait for the operational savings afforded by robotic piece-picking.
Reliability of the system
Robotic piece-picking is reliable if it provides order integrity – always picking the right quantity, and successfully transferring it to the place location. Robots can also provide and contribute to order integrity by validating an item, such as via barcode scan or image matching, verifying that the proper inventory was presented to the piece-picking system. Similarly, robots can capture data regarding picking tasks, including images, that can be used to confirm task completion, improving overall operational reliability and contributing to continuous improvement for your key quality metrics.
Robotic piece-picking also must achieve a high degree of robot independence, operating with minimal human intervention, in order to satisfy use requirements and maximize benefits for warehouse processes. Robot independence can be achieved by having high operational reliability, incorporating error and exception handling into the software control systems, and providing a simple, robust mission control protocol for interfacing to host systems in the facility.
Robotic systems must be mechanically reliable. They should leverage highly reliable COTS technology and subsystems that are supported by data from production use at scale. For newer elements, such as gripper hardware, quick-swap designs and on-site spares programs help sustain continuous operation.
In addition, a successful robotic piece-picking solution must be reliable even in the case of exceptions or errors. It’s easy to plan for happy-path scenarios, where everything is perfect. But this ignores the fact that even in the best-run facility or with the best robotic piece-picking workcell design, there can be errors. A well-conceived solution will have a simple message interface that can easily integrate with the warehouse control system (WCS) to coordinate and resolve issues – and the issue should resolve automatically, which requires robust error handling in the integration message protocol.
Canadian pilot program
An early example pilot project involved a Canadian e-commerce company evaluating robotic piece-picking for the ASRS pick-tending use case.
The system was deployed and integrated with the existing WCS software and physically at the ASRS system port. The task included picking from different bins in sub-divided totes and placing to the proper outbound carton or tote, for several orders that were in queue. The robotic system would receive pick missions from the WCS and would confirm mission success after placing the item in the appropriate container. In the event of an error, including empty source tote, misaligned tote dividers, or a missed pick, for example, the system would log and annotate the error, message the WCS with an error code and then the workflow would proceed based on a predefined set of error handling rules.
The robotic system was initially able to pick over 70 per cent of the SKU population with no modification to how the items were slotted in the storage totes and with no modifications to packaging. Keeping in mind that not all SKUs represent an equal proportion of volume, this was encouraging. Depending on the product details, robotic systems may be able to pick 90 per cent or more of the overall SKU set and an even higher per cent of volume, for products that are managed in ASRS.
An interesting example of a product that can be hard for robots to pick is men’s belts. If they arrive at the facility ready to hang on racks in a store, it’s not a great form factor for robotic perception and grasping systems. Rolling the belts and securing them in a snug plastic bag secured with an elastic makes them easier to pick and also prepares them for pack-out.
Raw rates for the robotic system were in the 400-500 uph range, including delay times in messaging between the systems – which can be reduced in future work – and time to reach to the farthest outbound container. This range is fast enough to where the throughput is now limited by the upstream and downstream systems. It could be further optimized in stations designed with robot ergonomics in mind.
Reliability improved throughout the trial with attention being paid to how some items were arranged in the bins. If bagged t-shirts were packed tightly and arranged vertically in the totes, it would make it a two-hand pick for a person, as one hand separated the items and the other grabbed an item. For the robotic system, laying the items on their side enabled successful single gripper picks. These and other “lessons learned” are driving new feature development based on capabilities of the robotic vision system and machine learning, as well as adaptive hardware systems.
In this example project and others, suppliers and customers are quickly learning how to effectively leverage the capabilities of robotic piece-picking to tackle the challenges in the dynamic world of e-commerce.
Robotic piece-picking at work
Robotic piece-picking is automation in action. The basic value framework is driven by the 3Rs of range, rate and reliability as they apply to your business, and best-in-class solutions have a clear set of common characteristics that offer high levels of 3R performance. When these new and complex technologies are combined intelligently, the results are simple: items are picked and placed predictably, and customers receive their orders on time.
Vince Martinelli is the head of product at RightHand Robotics, a leader in providing robotic piece-picking solutions that improve performance and efficiency in e-commerce order fulfillment and intralogistics.
This article originally appeared in the October 2018 issue of Manufacturing AUTOMATION.